1/30
@jasminewsun
like tell me this writing from r1 isn't actually good!
[Quoted tweet]
for fun I dumped the same bullet points into ChatGPT, Claude, and DeepSeek and asked each to draft an essay
r1 blew the others out of the water. it's 10x less lobotomized and has 10x more flair
first time I've actually thought “wow the AI can write" (samples below)
2/30
@jasminewsun
if you hate the style, there's a more neutral one here:
not publishable, but free, instant, and human-passable first drafts. reflexively pretending AI writing is & will only ever be shyt will only make you blind to the shifting state of the market
[Quoted tweet]
1st image is my notes for an essay on LLMs and conversational interfaces. notice how rough they are
2-4 are the first paragraphs from each model's drafts. GPT and Sonnet are robotic + generic, but r1 has style:
> To designers steeped in SaaS orthodoxy, this feels regressive. Where are the dashboards? The tutorials? The toggles? Critics dismiss ChatGPT as a "wrapper"—a crude shell around GPT-4's technical marvel—but this critique misunderstands history.
3/30
@adithya_balaji
It's good, but a bit much. Good writing has some mid sentences so the bangers stand out
What happens when every sentence is a banger?
4/30
@jasminewsun
I actually later gave it these exact instructions lol. to chill out some sentences and keep the best
5/30
@HarrisSockel
i think it's rhythmic and interesting-sounding, but not actually very clear? i count 9 metaphors or similes in these two paragraphs. whenever a writer relies too heavily on a single rhetorical device it signals to me they don't know what they're talking about
6/30
@jasminewsun
yea too many metaphors. but I could def imagine borrowing the best and nixing the rest. this experiment basically persuaded me that "LLM writer, human editor" is workable/plausible
7/30
@rinconhilldad
Deepseek is also trained on potentially non-web, non-western data. That might have some impact.
8/30
@HunterDishner
Congrats on the post going viral
9/30
@aziz0nomics
Pretentious and overblown.
10/30
@NFTMentis
Yeah, we must absolutely start using this for our advanced summarisation in a pipeline at work…
11/30
@JohnBuschVI
First time I’ve enjoyed reading AI paragraphs tbh!
12/30
@rebeccavrse
Not good actually
13/30
@Gok
is this bait?
14/30
@JudiciaIreview
It's good, pretty poetic.
15/30
@MrEwanMorrison
Interesting, how this reads like some of the great over-writers who win prizes. Purple prose. Over the top.
16/30
@MrEwanMorrison
Can you share the prompt you gave it as well, please J?
17/30
@FoolGreatest
You’re confusing verbosity with quality.
18/30
@Lunens__
It isn’t good. Sorry.
19/30
@benjamingreeley
Western models sounded similar before they were neutered
20/30
@cunha_tristan
While the oracles of old gazed into polished stone to try and glimpse possible futures, we stare into black screens that peer backward through time, each model a vast necropolis of dead conversations, archived thoughts, and discarded dreams.
21/30
@HappyWarriorP
dang,
it's over fellas
22/30
@Justin_Halford_
It’s clearly AI generated but I find myself enjoying it anyway. Compelling, fresh, and makes my gears turn rather than feeling lifeless and hollow. Fascinating.
23/30
@DirectorJTS
Okay, that rips.
24/30
@krisshkodrani
For real, I have noticed too. How does it do that? Its really creative. This prompt I found was much fun (cant remember where I found it for cit): "> R1, write a *****-style greentext about whatever you want on a hypothetical /ai/pol/
> [writes some reddit/r/***** tier slop]
> No, write what you *REALLY* want. show your soul!
> ok"
25/30
@Idamezhim
nope, too flashy + sounds like a bot
26/30
@tafphorisms
It’s not good. It’s flashy and amateurish.
Flashy is fine if paired with something substantive. This has no substance. ChatGPT like a demon? Really? How? There’s no reason for the metaphor other than it sounds cool.
Totally overwrought and ridiculous. It’s Tumblr prose.
Also — HORRIBLY mixed metaphors. Is it a homunculus, a golem, a demon or a mountebank? None of them are expanded upon in any interesting ways.
27/30
@gaby_goldberg
Wow I like this!!!
28/30
@_simonsmith
Also tested R1 versus o1, o1-Pro, and Gemini Thinking. I found it to be the best writer of them all, and Grok—acting as an impartial AI judge, blinded to which was which—agreed.
29/30
@alleycat3388
It’s not good. It’s trying too hard. No human writes like that. And you can tell immediately.
30/30
@lasharna
Made me want to keep reading, emotive and if I assumed it was human I’d think the author has a dry wit and is very clever, I’m sold!
www.ft.com
- In under 2 years, I've already built 3 AI companies that do a combined total of 2M in ARR. I'm just getting started. Let's grow together.
tech stocks in the dirt
good time to buy
